Reconstruction of Weight Matrices in Neural Networks : A Method of Correlating Outputs with Inputs

Abstract

Introducing a forgetting process in the learning phase of the neural network resulted in localizing the connections in the weight matrices. Such weight matrices, reconstructed weight matrices, indicated the active neurons in the network and were used to analyze the relationships between the input and output data. The relationship between the adaptive least-squares (ALS) method and the neural network led to the conclusion that the operation of the ALS method is, indeed, a restricted operation of the neural network.